Finding the stories within the page.

Preservation makes discovery possible.

The Library of Congress and its partners have preserved and digitized newspapers at an extraordinary scale. The original page is essential evidence: type, photographs, advertisements and neighboring stories all carry meaning.

But a broadsheet is a crowded reading surface. Small print, multiple columns and story continuations make it difficult to follow one article. OCR adds a searchable transcription, but a misread name or a heading separated from its body can hide the very story a reader needs.

Ink adds a reading layer to that work. It does not replace the scan or the institution that provides it.

From blocks to stories.

A block is a source region and its extracted text. A group connects regions that appear to belong together. A group can contain a headline, subhead and body; it can also contain a notice, advertisement or other material.

Article search presents stored groups with a headline and enough body text to be useful. This is a structural filter, not a verdict that the story is complete or correctly classified. Isolated fragments remain available under “Explore source blocks.”

Grouping uses the layout and OCR we already have: columns, position, type size and source relationships. We are improving and testing those rules before adding broader AI classification or rerunning expensive OCR.

TextSheet keeps the evidence close.

Read a page as columns of text. Within a group, switch between the contributing blocks, formatted text and the original crop. Swap an entire column to see its continuous scan. The full page remains available when context matters.

Search results open TextSheet at the matching source block. Publication, issue date, page and block location keep the passage connected to its origin.

Choose a way into the archive.

• Search articles for names, places and subjects, then refine by paper and year.
• Find a newspaper and check which dates are available.
• Browse an issue when you know the date, or search inside it in TextSheet.
• If a word is missing, try a spelling variant or shorter query. Explore source blocks or read the scan when article search cannot find it.

No result is not proof of absence. Coverage, OCR quality and grouping all affect what search can retrieve.

What we are testing next.

Our immediate focus is reliable headline and body attachment on a manageable set of papers. Photographs, captions, cross-column continuations and reviewed classification need further work. Better OCR and AI analysis are possible later stages, not guarantees attached to today’s results.

For archive partners and researchers, TextSheet makes the intermediate evidence inspectable: which regions formed a group, what text was extracted and where an attachment went wrong. That gives us concrete cases to evaluate together.

Ink is informed by work such as Harvard’s American Stories project. Our collections draw on several sources and are not all outputs of the same models. Source credit stays with each issue.